From: Amara D. Angelica (amara@kurzweilai.net)
Date: Tue May 06 2003 - 20:00:14 MDT
In Journal of the ACM (JACM), Volume 50 , Issue 1 (January 2003)
http://204.194.72.101/www/oy8guwod/Feigenbaum_notes.doc, Ed Feigenbaum
presents an interesting solution to this problem, proposing working
within formal, constrained knowledge domains and "distilling from the
WWW a huge knowledge base, reducing the cost of knowledge engineering by
many orders of magnitude."
He points out that "the WWW is not a knowledge base. Except for
supporting data, most of what the WWW contains can not participate
directly in inference making, problem solving, and decision making that
a CI must do. The WWW, simply put, does not represent knowledge using
any of the standard tools of knowledge engineering, logic, and
probabilistic inference."
He goes on to point out that the Semantic Web is the solution: the
creation of shared ontologies and the use of "a markup language in which
each [Web page creator] can do an extensive semantic markup of his/her
textual submission," such as the "DAML+OIL [and its successor, OWL]
international project has the promise of eventual distribution of
user-friendly semantic processing and markup tools to all web page
builders."
"Knowledge engineers must build a system of "semantics scrapers" that
will access the semantic markups, integrate them appropriately into the
growing knowledge base, and set up the material for the scrutiny of an
editorial process."
I would be interested in your opinion of this approach for generating
constrained knowledge domains. Would the path to a super AI then be via
some form of metaontology that could automagically integrate the various
domains?
> -----Original Message-----
> From: owner-sl4@sl4.org [mailto:owner-sl4@sl4.org] On Behalf
> Of Ben Goertzel
> Sent: Tuesday, May 06, 2003 6:21 PM
> To: sl4@sl4.org
> Subject: RE: Hardware Progress: $212/GFlop/s
>
>
>
> > I realize that there is a lot of subconscious information
> people have,
> > process, and use. But a super AI, given access to something
> like the
> > Internet, would probably have access to all of the raw
> information it
> > would need. If it was taught how to process that
> information, there'd
> > be little need to train it further - it could teach itself.
>
> That is quite true!!
>
> The question is, how will it be taught how to process that
> information? It will have to be taught this via guided
> experience, not (IMO) via being fed files of knowledge about
> how to process the information..... And this -- teaching how
> to process and learn from information -- is what some of us
> think may be a years-long teaching process...
>
> > Or maybe I am just naive. I just like reading this list, I haven't
> > much education as far as cognitive science goes :)
>
> Well, even if you ARE naive, that doesn't mean becoming an
> expert would make you agree with me!
>
> This is stuff on which experts disagree widely -- I have
> strong opinions, but others with equally many credentials
> have equally strong and strongly opposed opinions...
>
> And so it will remain till significantly more empirical
> progress is made in the AI domain!!
>
> ben g
>
>
>
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:42 MDT